Rapid Prediabetes Screening With a VOC Nanosensor: A Novel Saturation-Onset Prediction-Driven Classifier

Abstract

Prediabetes is a transitional metabolic state for which effective noninvasive screening tools are scarce. This letter proposes a novel knowledge-driven AI approach to rapidly classify prediabetic samples using a breath-based sensing system that integrates a chemi-resistive volatile organic compound nanosensor. This letter showcases two ML approaches and delivers inference utilising less than 1 min of sensor response. Model A: a feature-based classification approach for three class-diabetic, prediabetic, and nondiabetic samples with 90% accuracy, and Model B: directly does three-class classification by targeting early temporal responses for classifying three classes with 92% accuracy. The prediabetic class is detected with an F1-score of 0.96. This work demonstrates how embedded intelligence and adaptive sensing jointly enable early, reliable, and deployable point-of-care metabolic diagnostics.